Free Survey Bias Checker

Paste your survey questions, one per line, and see which ones are leading, loaded, double-barrelled, absolute, or hard to read. Each flag explains why the wording skews answers and how to fix it, and you can download the whole check as Word or PDF. It is free, needs no sign-up, and every question is checked locally in your browser. If you want a neutral rewrite of a flagged question, you can ask for one; that is the only step that sends anything to our server, and only for the question you pick.

Your questions

0 questions

One question per line

Nothing you paste here leaves this page. The check itself makes no AI call: it matches known biased-wording patterns locally in your browser. Asking for a rewrite later is a separate, optional step.

Report

SURVEY BIAS CHECK REPORT

Paste your survey questions on the left, one question per line, and the check runs as you type.

Nothing you paste leaves this page: every question is checked locally in your browser.

Generated by BlockSurvey

How the survey bias checker works

Paste your questions

Drop your survey questions into the box, one question per line. There is no limit, no sign-up, and the check starts as you type.

Read the flags

Each question is checked against nine biased-wording rules. Every flag names the pattern, its severity, why it skews answers, and how to fix it.

Download the report

Export the full check as a PDF or an editable Word file. The check runs locally in your browser, and the downloaded report is assembled on your device.

Free, private, and built for better research

Most question-review tools either paywall the result or upload your draft survey to someone else's server before you have decided to run it. This one does neither. The check is rule-based: nine documented patterns, matched against your text in the browser, with no model and no network call involved. Rewriting a flagged question is the one optional step that uses a server, and only ever for the single question you choose.

The check is in-browser

The matching is deterministic and there is no AI call behind it. Your questions are checked locally in your browser. Only the optional rewrite sends a question to our server, one at a time, and nothing is stored.

No account required

No sign-up, no email wall, and no record of the survey you are drafting.

Detection, honestly described

It flags biased wording and tells you how to fix each pattern. It does not rewrite your questions, because that needs a language model and this tool does not use one.

What it checks

Nine wording patterns, each with a severity and a fix:

Double-barrelled (high). Two topics joined by "and" or "or" with only one answer allowed. Split it into two questions.
Leading (high). Phrasing that signals the expected answer, such as "don't you agree" or "everyone knows". Ask neutrally and offer both directions.
Loaded or charged wording (high). Emotive words such as "amazing", "terrible", or "useless". Replace them with neutral description.
Assumptive or presupposing (high). Wording that assumes the experience happened or was enjoyed. Ask whether first, then how much, and add "not applicable".
Absolute wording (medium). "Always", "never", "every", "nobody". Use a frequency scale instead of forcing a false choice.
Double negative (medium). Two negations in one sentence. Rephrase positively so there is one possible reading.
Jargon and reading load (medium). Over 25 words, or several very long words. Shorten it and use plain language.
Possibly unbalanced scale (medium). Only positive anchors, such as "how satisfied" or "how good". Check the scale has a real negative end and a neutral midpoint.
Vague quantifier (low). "Often", "regularly", "several". Use concrete ranges such as "1-2 times per week".

Accuracy note: this is a fast, structured starting point, not an authoritative verdict. The checker matches known wording patterns, so it will miss bias that depends on context, such as a skewed answer scale, a loaded topic, or an assumption carried by the questions around it. It will also flag wording that is perfectly fine in context. Human judgement decides what to change.

Worked examples: biased vs neutral

The tool flags the pattern; you write the replacement. These examples, written by our research team, show what each fix looks like in practice.

Biased question	The problem	A neutral version
"Don't you agree that our friendly support team did a great job?"	Leading and loaded. It names the answer you want, twice.	"How would you rate our support team?" with a balanced scale from very poor to very good.
"How satisfied were you with the speed and price of delivery?"	Double-barrelled and assumptive. Fast but expensive has no answer, and it assumes satisfaction.	Two questions: "How would you rate delivery speed?" and "How would you rate delivery price?", each with a "did not use" option.
"How often do you use our amazing new dashboard?"	Loaded wording plus a vague quantifier. "Often" means different things to different people.	"In the last 30 days, how many days did you open the dashboard?" with the options 0, 1-3, 4-10, and 11 or more.
"Would you not be unwilling to recommend us to a colleague?"	Double negative. Answers measure reading comprehension, not intent.	"How likely are you to recommend us to a colleague?" on a 0 to 10 scale.
"Everyone knows remote work is better; do you always work from home?"	Leading premise plus an absolute. There is no honest answer for a hybrid worker.	"In a typical week, how many days do you work from home?" with the options 0 through 5.
"How much did you enjoy your favourite feature in our best-in-class platform?"	Assumptive and loaded. It presumes a favourite exists and that it was enjoyed.	"Which feature did you use most?" followed by "How would you rate that feature?", with "none of these" available.

Built for every team writing survey questions

Wording problems cost almost nothing to fix before a survey goes out, and there is no fixing them once fieldwork closes. Run the draft through first, then pair it with an AI survey platform when the answers you collect are sensitive.

Researchers & data teams

Catch leading and double-barrelled items before they contaminate a dataset you cannot re-collect.

Product & UX teams

Stop writing questions that flatter the feature you just shipped.

Marketing & customer experience

Get NPS and satisfaction scores you can defend when someone asks how the question was worded.

HR & people teams

Keep engagement surveys neutral, so staff answer what they think rather than what is expected.

Academics & students

Run a wording pass before an ethics review or a supervisor does it for you.

Teams adopting AI drafting

AI-drafted questionnaires inherit the flattery in their prompts. Check the output before it ships.

Now protect the answers you collect.

BlockSurvey is an AI survey platform, encrypted end to end, so the responses you gather are never sold, mined, or used to train models.

More free AI tools

Explore more

Survey Drop-off Estimator

Estimate how long your survey takes and where respondents are most likely to abandon it, before you send it.

Survey Response Quality Checker

Spot straight-lining, speeding, gibberish, and duplicate responses in your survey data, all in your browser.

AI Excel Formula Generator

Describe what you need in plain language and get the exact Excel or Google Sheets formula, ready to paste.

AI Persona Generator

Turn your research notes into a clear, shareable user persona you can download as Word or PDF.

AI Ethics Policy Generator

Draft an AI ethics policy your organization can publish, aligned with NIST AI RMF 1.0 and ISO/IEC 42001.

AI Acceptable-Use Policy Generator

Generate an AI acceptable-use policy naming approved tools, prohibited data, and guardrails.

AI DPA / Contract Addendum Generator

Create a signable AI data processing addendum aligned with GDPR Article 28.

AI Incident Response Plan Generator

Build an incident response plan for AI-specific failures like data leakage and model errors.

AI Model Card / System Card Generator

Document your AI system as a model card in Word, PDF, or Markdown.

AI Procurement Clause Generator

Generate ready-to-paste AI compliance clauses for procurement contracts and RFPs.

AI Disclosure Notice Generator

Write a made-with-AI disclosure notice aligned with EU AI Act Article 50.

AI Risk Assessment

Assess the risks of an AI system across data, bias, safety, and oversight, and download it as Word or PDF.

AI Governance Maturity Scorecard

Get a maturity scorecard of your AI governance with priority next steps, as Word or PDF.

ISO 42001 Readiness Assessment

Check how ready you are for ISO/IEC 42001 across the management-system areas, with a roadmap.

Frequently asked questions

What is the survey bias checker and what does it do?

It is a free tool that reads the survey questions you paste in and flags the wording patterns known to skew answers. It checks each line against nine rules: double-barrelled, leading, loaded wording, absolutes, double negatives, jargon and reading load, assumptive phrasing, vague quantifiers, and scales anchored only on the positive side. For each flag you get the severity, why that pattern distorts answers, and how to fix it. The whole check runs in your browser as you type.

How accurate are the results?

It gives you a fast, structured starting point, not an authoritative verdict. The tool matches known wording patterns, so it will miss bias that depends on context, such as a biased answer scale, a loaded topic, or an assumption carried by the questions around it. It will also sometimes flag wording that is perfectly fine in context. Read every flag and use your own judgement before you change anything.

Does this rewrite my questions for me?

It can, if you ask it to. The check itself is rule-based pattern matching that runs locally in your browser and does not rewrite anything: for each flag you get guidance derived from the rule that fired, such as "split this into two separate questions, one per topic" for a double-barrelled question. Alongside that, each flagged question has a "Suggest a rewrite" button. Rewriting a sentence well needs a language model, so that button sends the single question you picked to our server and returns a neutral version. It is optional, it runs one question at a time, and nothing you send is stored, sold, or used for training. Treat the result as a draft to review rather than a finished question.

What happens to the data I enter?

The check itself sends nothing anywhere. There is no AI processing and no API call behind it: your questions are matched against the rules locally in your browser, and the report you download is assembled on your own machine, which is why you can safely check a survey you have not launched yet. There is one exception, and it only happens when you click it: if you ask for a rewrite of a flagged question, that single question is sent to our server so a language model can rewrite it. It is processed to generate the rewrite and is never stored, sold, or used for training. Everything else stays on your device. Avoid asking for a rewrite of a question containing confidential or personal information.

Is it free? Do I need an account?

It is completely free and needs no account. There is no sign-up, no email wall, and no watermark. You get the full report as an editable Word file or a PDF, not a preview of a paid version.

What is a double-barrelled question?

A double-barrelled question asks about two things at once but allows only one answer. "How satisfied were you with the speed and price of delivery?" is the classic shape: someone who loved the speed and hated the price has no honest answer to give, and you cannot tell from their response which half they were rating. The fix is always the same: split it into two questions, one per topic.

What is a leading question?

A leading question signals the answer it expects, so respondents drift towards it. "Don't you agree that our support team is doing a great job?" tells the reader which side is the agreeable one, and agreement rates rise for reasons that have nothing to do with the support team. Neutral phrasing presents both directions as equally acceptable: ask how someone would rate the support team, and let a balanced scale record the answer.